| Network Working Group | M. Bhatia |
| Internet-Draft | Alcatel-Lucent |
| Intended status: Standards Track | M. Chen |
| Expires: July 07, 2012 | Z. Wang |
| Huawei Technologies Co., Ltd | |
| L. Guo | |
| China Telecom | |
| M. Binderberger | |
| January 6, 2012 |
Bidirectional Forwarding Detection (BFD) on Link Aggregation Group (LAG) Interfaces
draft-mmm-bfd-on-lags-02
This document proposes a mechanism to run BFD on Link Aggregation Group (LAG) interfaces. It does so by running an independent BFD session on every LAG member link.
(For IP/UDP encapsulation)
A dedicated well-known multicast IP address for both IPv4 and IPv6 is introduced as the destination IP address of the BFD packets when running BFD on the member links of the LAG.
(For Ethernet encapsulation)
A new Ethernet type is introduced to send BFD packets directly in Ethernet frames when running BFD on the member links of the LAG.
There is currently also no standard that describes how BFD runs on a LAG interface as a whole. This draft proposes a definition for this problem too while taking into consideration existing implementations.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119].
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at http:/⁠/⁠datatracker.ietf.org/⁠drafts/⁠current/⁠.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on July 07, 2012.
Copyright (c) 2012 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http:/⁠/⁠trustee.ietf.org/⁠license-⁠info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
The Bidirectional Forwarding Detection (BFD) protocol [RFC5880] provides a mechanism to detect faults in the bidirectional path between two forwarding engines, including interfaces, data link(s), and to the extent possible the forwarding engines themselves, with potentially very low latency.
BFD can be used for detecting failures of the path between two network devices. Typically the application clients are not aware of any inner structure of the underlying interface, being layer 3 applications themselves like Open Shortest Path First (OSPF) [RFC2328] or Border Gateway Protocol (BGP)[RFC4271]. While this works for interfaces like Ethernet and Packet Over SONET (POS), it causes problems for bundled interfaces like LAG.
A LAG is used to bind together several physical ports between two adjacent nodes so they appear to higher-layer protocols as a single, higher bandwidth "virtual" pipe. A LAG interface thereby allows aggregation of multiple network interfaces as one virtual interface for the purpose of providing fault-tolerance and higher bandwidth.
The problem with running BFD over a LAG is that with a single BFD session and without internal knowledge of the LAG structure it is impossible for BFD to guarantee a detection of anything but a full LAG shutdown within the BFD timeout period. The LAG shutdown is typically initiated by some LAG module, which we will refer to as the LAG Management Module (LMM) in the rest of the document. LAG timers are typically multiple times slower than the BFD detection timers (multiple 100msec of LMM vs. multiple 10msec of BFD). There is thus a need to bring some sort of determinism in how BFD runs over a LAG. There is also a need to detect member link failures much faster than what Link Aggregation Control Protocol (LACP) allows.
The document proposes establishing a BFD session over every member link the LAG is built upon. BFD can combine these information to provide fast detection for layer-3 applications.
While there are native Ethernet mechanisms to detect failures (802.1ax, .3ah) that could be used for LAG, the solution proposed in this document enables operators who have already deployed BFD over different technologies (e.g. IP, MPLS) to use a common failure detection mechanism.
The simplest approach to run BFD on a LAG interface is to ignore the internal structure and treat the LAG as one "big pipe". We call this mode of operation as "BFD over Big Pipe" or "BBP" for short. It corresponds to section 7.1 in RFC 5882 [RFC5882].
We need to standardize the BBP approach. The following requirements define what it means to treat a LAG interface as a single interface with no additional structure:
The BFD session on the LAG interface then follows RFC 5880 and RFC 5881 in all details.
Because there is no standard, vendors have implemented their own proprietary mechanisms to run BFD over LAG interfaces. Two examples are shown here. Both satisfy the requirements in Section 2.1
Some implementations send BFD packets only over one member link. Others spray BFD packets over all member links of the LAG. There are issues with each of these approaches.
In the first approach, BFD sends packets onto the LAG and the LAG load balance algorithm will select a member port, which may be the same port for all the packets of this BFD session. BFD will remain up as long as this "primary" port is alive. It will go down once the primary port goes down till another port is selected as the primary. Problems arise with this design as BFD is oblivious to the presence of other member links in the LAG. If a non-primary link goes down, the BFD session remains unaffected as it can still send and receive BFD packets over the primary link. This results in all traffic sent over the failed member link getting dropped, till the LMM removes the failed link from the LAG. Conversely if the primary link goes down, then the BFD session will go down, till a new member link is elected as the primary link.
In the second approach, BFD packets are sprayed over all the member links of a LAG. This is done naively via round-robin, where each BFD packet is sent using the subsequent member link, in a round-robin fashion. It solves the problem of BFD going down because of the primary port going down, but it still does not solve the problem of traffic getting lost when one of the member link goes down. This is because, when a member link goes down, BFD remains up and traffic continues to go over the link that has failed till a higher layer protocol detects this and removes the offending link from the LAG.
Between the two approaches the second one is RECOMMENDED as its much more flexible and is not prone to single link failures. To completely solve all issues we RECOMMEND running BFD on all member links as described in Section 3.
The mechanism proposed for a fast detection of LAG member link failure is running BFD sessions on every LAG member link. We name this mode of BFD operations "BFD on LAG members" or "BLM" for short. It corresponds to section 7.3 in RFC 5882 [RFC5882].
The overall BLM session consists of the LAG interface, i.e. the aggregated link, a set of BFD sessions running on the member links and a new BFD state for the LAG; this state is explained in more detail in Section 3.3. We call the member-link sessions as micro BFD sessions; their details are discussed in Section 3.2.
The set of micro sessions is such that we have one micro session per member link. This set can change over the lifetime of a BLM session. E.g. BFD receives updates for the micro session set when links are physically added or removed from the LAG and will accordingly create or delete micro BFD sessions.
The details of how the update happens are implementation specific and outside the document's scope. For example the client requesting the BLM session could provide these updates.
(The following paragraph applies only when IP/UDP encapsulation is in use) Only one address family MUST be used per BLM session, i.e. the set of micro BFD sessions belonging to the BLM session MUST either all use IPv4 or all use IPv6.
Multiple BLM session requests for the same LAG interface result in a shared BLM session. The set of micro sessions finally used is the superset of the individual micro session sets. If conflicting session parameters are requested then it is a local issue as to how to resolve the parameter conflicts, as explained in RFC 5882, Section 2.
A single micro BFD session runs on every member link of the LAG. These micro BFD sessions follow RFC 5880 [RFC5880].
Only asynchronous mode is considered in this document. The echo function is outside the document's scope. At least one system MUST take the Active role (possibly both). The micro BFD sessions on the member links are independent BFD sessions. They use their own unique, local discriminator values, maintain their own set of state variables and have their own independent state machine. Timer values MAY be different, even among the micro sessions belonging to the same LAG, although it is expected that micro sessions belonging to the same LAG use the same timer values.
The demultiplexing of a received packet is solely based on the Your Discriminator field, if this field is nonzero. For the initial Down packet of a micro session this value may be zero. In this case demultiplexing MUST be based on some combination of other fields which MUST include the interface information of the member link.
When receiving a BFD packet for a micro session with a valid, non-zero Your Discriminator then a check MUST be done if the packet was received on the correct member link interface. If the check fails then the packet MUST be discarded. This test needs to be done before state variables for the micro sessions are updated by the received packet.
[Either this section or the alternative sections Section 3.2.3, Section 3.2.2 should remain in the final document. There is no intention to support multiple encapsulations.]
The BFD Control packets for each micro BFD session are IP/UDP encapsulated as defined in [RFC5881]. They use a well-known link-local multicast IP address (224.0.0.X for IPv4, FF02::X for IPv6, to be assigned by IANA).
On Ethernet-based LAG member links the corresponding destination multicast MACs will be 01:00:5e:00:00:XX for IPv4 and 33:33:00:00:00:XX for IPv6. Each member link uses its own MAC address as the source MAC address.
[Either this section or the previous sections Section 3.2.3, Section 3.2.1 should remain in the final document. There is no intention to support multiple encapsulations.]
The BFD Control packets for each micro BFD session are IP/UDP encapsulated as defined in [RFC5881], but with one major change: the UDP destination port will not be 3784 but "BfdBndlPort" (to be assigned by IANA). Control packets use a destination IP address that is the peer's remote IP address. The details of how this destination IP address is learnt is beyond the scope of this document.
On Ethernet-based LAG member links the destination MAC is the MAC assigned to the peers LAG aggregator.
[Either this section or the next sections Section 3.2.1, Section 3.2.2 should remain in the final document. There is no intention to support multiple encapsulations.]
The BFD packet is directly encapsulated into the Ethernet frame. The frame has the following format: Ethernet header according to [IEEE802.3], then Type/Length field set to "BfdEtherType", followed by the BFD packet
The Ethernet payload must be padded with zeros to reach 46 bytes if the BFD packet size is not already larger.
When receiving an Ethernet frame the payload is used for further BFD processing. Additional padding data MUST be ignored if it was required to reach the minimum payload length of 46 bytes.
IANA needs to assign a L2 MAC address according to [RFC5342] that would be used as the destination MAC for all control packets in the micro BFD sessions.
A new Ethertype must be assigned by the IEEE Registration Authority to the BFD over Ethernet protocol that will be used for all micro BFD sessions.
An additional state variable is introduced for BFD on LAG members: the concluded state. The state values are Down, Up and AdminDown. This state is not part of the micro session state machine. Instead it describes the overall state of the LAG. It is a local state and does not appear (directly) in any BFD packet on any link.
The concluded state may be set to AdminDown for administrative purpose, to keep the BLM and the micro sessions indefinitely down. When the concluded state is entering AdminDown then all micro sessions belonging to the BLM MUST enter the AdminDown state as well.
A function must be defined, which evaluates all the states of the micro sessions that belong to the BLM. This function has two output values Down and Up and the concluded state is updated with the last evaluation result, unless it is already in AdminDown state. The evaluation takes place whenever a micro session is added, removed or is changing state.
The details of the evaluation function are outside the scope of the document. The function could for example test for a minimum number of micro sessions in Up state. The function could even be "outsourced" and e.g. the decision logic of the LMM module could be used.
The concluded state is important for layer-3 clients requesting BFD sessions over the LAG or over Vlans on the LAG. Details will be discussed in Section 4.
[Either this section or the next section Section 3.4.2 should remain in the final document. There is no intention to support both encapsulations.]
The user interface for BFD micro sessions encapsulated in IP/UDP MUST allow to send an IP/UDP packet on a specified LAG port. When receiving BFD packets for micro sessions then the IP/UDP packet MUST be provided together with an information what LAG port the packet was received on.
[Either this section or the previous section Section 3.4.1 should remain in the final document. There is no intention to support both encapsulations.]
The user interface for BFD directly transported in an Ethernet frame MUST allow to send and receive a complete Ethernet frame with the specific "BfdEtherType" type value. The information that specifies the LAG port from which a frame is sent to or received from is either an explicit extension of this API or is implicitly given by binding the API to a specific port.
As an example, the API could be identical with the MA_DATA request and indication as defined in section 2.3 of [IEEE802.3].
Layer 3 protocols like e.g. OSPF may use BFD on LAG members in one of the following ways:
This state update allows BBP session to run with more relaxed timer values as the more intense liveliness detection is done by the micro BFD sessions.
An implementation MUST provide a configuration knob which lets the user select the mode.
There are certainly many ways to use BLM. Here is one example envisioned by the authors.
The LAG Management Module (LMM) could be envisaged as a client of BFD, i.e. the LMM requests a BLM session and takes responsibility to update the set of micro sessions
LMM then uses BFD, instead of or in parallel with LACP, to monitor the health of the individual members links of the LAG. Details are outlined below.
Bringing a member link up:
When the status of a port is about to change to Distribution TRUE (see section 5.3.15 in [IEEE802.1AX]), i.e. before the port is added to the distribution function of the LAG, then the particular BFD micro session is requested. An implementation MAY wait for the micro BFD session to reach Up state before adding the port to the LAG's distribution function and changing the port status to Distribution TRUE.
In case LACP is in use then the steps of the previous paragraph are executed in the "Distributing" state (see the Mux machine state diagram Figure 5-14 in [IEEE802.1AX]). I.e. LACP is in Distributing state before the implementation potentially waits for the BFD micro session to reach Up state.
Detecting a member link failure:
When running in parallel operation the logic for failure is that both LACP and BFD can indicate a failure.
When a micro BFD session, that runs on a member link of a LAG, goes down then this member link MUST be taken out of the distribution function of the particular LAG and the port status MUST change to Distribution FALSE. The BFD micro session for the link MUST be deleted when the link has been taken out of distribution.
In case LACP is in use then the variable "Selected" MUST be set to UNSELECTED when BFD reports a Down state. The steps of the previous paragraph are executed in the "Collecting" state (see the Mux machine state diagram Figure 5-14 in 802.1AX).
The behaviour of the LMM MUST be configurable if waiting for a BFD status of Up to add a member link is supported, to allow an alternative mode of adding the member link irrespective of the BFD state for interoperability purpose. Bringing the member link up without waiting for BFD is then the default behaviour.
This document does not introduce any additional security issues and the security mechanisms defined in [RFC5880] apply in this document.
Routers compliant to this standard will now need to process packets addressed to a new multicast address. This however, should not open any new attack vector as it is a link local multicast and the attacker would have to be on the same link as the router to launch such packets.
This document does not introduce any additional security issues and the security mechanisms defined in [RFC5880] apply in this document.
If no mechanism exists to transport Ethernet frames from a node other than a directly connected node then the security is identical to the TTL=255 check for IP packets.
The IANA is requested to assign a well-known link-local multicast IP address: "224.0.0.XXX" for IPv4 and FF02::X for IPv6.
The IANA is requested to assign a well-known port number for the UDP encapsulated micro BFD sessions.
IANA needs to assign a L2 MAC address according to RFC 5342 [RFC5342] that would be used as the destination MAC for all control packets in the micro BFD sessions.
(The following applies only in case of Ethernet encpasulation) A new Ethertype must be assigned by the IEEE Registration Authority to the BFD over Ethernet protocol that will be used for all micro BFD sessions.
Most of the text for this document came originally from draft-chen-bfd-interface-00.
We would like to thank Dave Katz, Alexander Vainshtein, Greg Mirsky and Jeff Tantsura for their comments on this draft.
We would also like to thank the members of the BFD WG who expressed strong support about the need to run BFD on all the member links of a LAG.
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
| [RFC5880] | Katz, D. and D. Ward, "Bidirectional Forwarding Detection (BFD)", RFC 5880, June 2010. |
| [RFC5881] | Katz, D. and D. Ward, "Bidirectional Forwarding Detection (BFD) for IPv4 and IPv6 (Single Hop)", RFC 5881, June 2010. |
| [RFC5882] | Katz, D. and D. Ward, "Generic Application of Bidirectional Forwarding Detection (BFD)", RFC 5882, June 2010. |
| [RFC0768] | Postel, J., "User Datagram Protocol", STD 6, RFC 768, August 1980. |
| [RFC0791] | Postel, J., "Internet Protocol", STD 5, RFC 791, September 1981. |
| [RFC2328] | Moy, J., "OSPF Version 2", STD 54, RFC 2328, April 1998. |
| [RFC4271] | Rekhter, Y., Li, T. and S. Hares, "A Border Gateway Protocol 4 (BGP-4)", RFC 4271, January 2006. |
| [RFC5342] | Eastlake, D., "IANA Considerations and IETF Protocol Usage for IEEE 802 Parameters", BCP 141, RFC 5342, September 2008. |
| [IEEE802.1AX] | IEEE Std. 802.1AX, "IEEE Standard for Local and metropolitan area networks - Link Aggregation ", November 2008. |
| [IEEE802.3] | IEEE Std. 802.3, "IEEE Standard for Information technology - Telecommunications and information exchange between systems - Local and metropolitan area networks - Specific requirements Part 3: Carrier sense multiple access with collision detection (CSMA/CD) access method and physical layer specifications ", December 2008. |
[This section will finally go away. It documents some of the discussions and decisions made recently on the BFD mailing list.]
The destination IP address for the BFD control packets for the micro BFD sessions can be Unicast or Multicast. Each has its set of advantages and disadvantages.
Advantages with using a Unicast IP destination address:
Disadvantages with using a Unicast IP destination address:
Advantages with using a Multicast IP destination address:
Disadvantages with using a Multicast IP destination address:
Based on the above analysis, we decided to go with multicast IP addressing scheme for the micro BFD sessions.
While we personally think that the Multicast solution for micro BFD sessions is better then the Unicast, we briefly describe how we could make Unicast work.
Once LACP has brought up the links, routers will initiate establishing a Unicast BFD session over each component link of the LAG. The remote destination addresses could either be configured on the routers or could be discovered via some discovery protocol (that can be standardized later). The exact mechanism to get the destination IP address is beyond the scope of this document.
Some service providers have expressed interest to run BBP on top of the micro BFD sessions. In this case, its imperative that Unicast BFD packets corresponding to the micro sessions use a different UDP port (assigned by IANA) lest they get mixed up with the BFD packets meant for the BBP sessions.
This design requires LACP to be present so that it brings up the links and ARP processing can begin. Operators however have also expressed interest in a solution that works in the absence of LACP. This could be done by using a well known L2 MAC address to carry the micro session BFD packets. This way routers dont have to depend upon ARP to boot strap the micro BFD sessions.
With at least three implementations using IP/UDP for the BFD packet encapsulation on the LAG member links there cannot be any doubt that technically IP/UDP encapsulation works for this purpose. What such a view is missing though is the requirement to have some kind of standardized packet send and receive API to allow everyone to implement the new standard.
The user interface for IP/UDP packets would be either for UDP, defined in RFC 0768 [RFC0768], or the IP user interface, defined in RFC 0791 [RFC0791]. None of them allows to provide any control about the LAG member port a packet is transmitted nor does it provide the information on which port the packet was received. Thus an agreement is required to extend these APIs to control the sending port and to know about the receiving port.
If we don't use the layer-3 user interface then we need to look at 802.3 and 802.1AX standards, as they describe in the case of a LAG what is "below" the IP layer. In this case we are already on the Ethernet layer and adding IP and UDP headers to the BFD packet may either conflict with IP/UDP itself or may be without any function. Thus encapsulating BFD directly in Ethernet and using a user interface fitting into 802.3 and 802.1AX seems a viable approach.
An additional sublayer is inserted between the MAC or MAC control of the physical port(s) and the Link aggregation sublayer. This allows to receive and inject BFD packets on every LAG port.
This additional sublayer allows to drop Ethernet frames with a specific Ethernet type (we name the value "BfdEtherType" from now) off the stream of frames coming from the MAC layer. It hands over the dropped-off frames to the BFD module. The new sublayer also allows to inject Ethernet frames with the specific Ethernet type into the stream of frames towards the MAC layer. All other frames are passing transparently between the MAC and the link aggregation layer.
+------------+
| Mac Client |
+------------+
^ |
| |
...................
| |
| V
+---------------------------------+
| Link aggregation |
| sublayer |
+---------------------------------+
^ ^
| |
........................................
| |
| | +-----+
| +---------- | ---------------->| BFD |
| | | +--------->| |
V (A) V (B) V V +-----+
+-------------+ +-------------+
| inject/ | | inject/ |
| drop-off | ... | drop-off |
+-------------+ +-------------+
^ (C) ^
| |
........................................
| |
V V
+-------------+ +-------------+
| MAC control | ... | MAC control |
| (optional) | | (optional) |
+-------------+ +-------------+
| MAC | | MAC |
+-------------+ +-------------+
| Physical | | Physical |
| layer | | layer |
+-------------+ +-------------+
Inject/drop-off mechanism for specific BFD Ethernet frames
Figure 1
The API in (A) behaves like the MAC side of the API defined in section 2.3 of [IEEE802.3]. All MA_DATA and MA_CONTROL requests are passed transparently to the API in (C), which behaves like the MAC Client side. Vice versa all MA_CONTROL indication received at (C) are passed transparently to (A). MA_DATA indication received at (C) are passed to (A) when the Ethernet Type is not BfdEtherType. Otherwise the MA_DATA indication is passed to API (B), which behaves like the MAC side of the API in section 2.3 [IEEE802.3] but without any MAC Control support.
A MA_DATA request received at (B) is passed to (C) if the Ethernet Type field in the frame is set to BfdEtherType; otherwise the frame is dropped.
When the BFD encapsulation is Ethernet then the following discussion is obsolete. In case of IP/UDP encapsulation it should be highlighted that the way a BLM session is defined above means a BLM request for a LAG with IPv4 and a BLM request for the same LAG with IPv6 is considered a shared session, with the obvious conflict that the micro session must be all either IPv4 exlcusiv-or IPv6. One could consider to allow BLM-v4 and BLM-v6 for the same LAG instead, which would mean we have to separate concluded states. This would require more details in Section 4.